NVIDIA’s Vera CPU — its first custom processor — is shipping now, and the deliveries read like a tour of the AI industry’s biggest names.
A CPU for the Agent Era
The insight behind Vera: agentic AI puts pressure on CPUs in ways traditional designs never anticipated. An agent’s sandbox, its tool calls, the orchestration layers, long-context state management — that’s all CPU work, concurrent and real-time. Vera’s answer:
- 88 custom NVIDIA-designed Olympus cores
- 1.2TB/s of memory bandwidth
- Up to 1.8x faster per-core performance on agentic AI workloads
As NVIDIA VP Ian Buck put it: “AI agents don’t run on GPUs alone… Agentic AI is creating a new CPU moment in the AI factory — as models move from answering to acting, Vera is purpose-built to keep that work moving at scale.”
Hand-Delivered Across the Ecosystem
The delivery tour is its own story: Buck personally handed Vera systems to Oracle Cloud Infrastructure (first cloud deploying at hyperscale — hundreds of thousands of Vera CPUs planned), Anthropic (head of compute James Bradbury), OpenAI (compute infrastructure head Sachin Katti), SpaceXAI (evaluating Vera for RL workloads and agent simulation, with Musk quizzing the team on cores, memory layout, and cooling), and most recently AWS — alongside an AWS-NVIDIA expansion that adds 2 million more NVIDIA GPUs including Vera Rubin GPU infrastructure.
Vera also serves as the host processor in Vera Rubin NVL72, paired with Rubin GPUs via NVLink-C2C in a unified memory architecture.
Read the full story at blogs.nvidia.com/blog/vera-cpu-delivery.